Journal of Personality and Social Psychology
1 W , Vol. 72, No. 3, 592-603
Copyright 1997 by the American Psychological Association, Inc.
0022-3514/97/S3.00
When Do Opposites Attract? Interpersonal
Complementarity Versus Similarity
D. Christopher Dryer and Leonard M. Horowitz
Stanford University
Two experiments examined whether interpersonal complementarity or similarity influences people's
satisfaction with dyadic interactions. Participants in complementary partnerships (submissive people
with dominant partners, dominant people with submissive partners) reported more satisfaction than
did those with similar partners. In Study 1 complementarity referred to the match between, the
participants' self-reported interpersonal style (dominant or submissive) and the role enacted by a
confederate (dominant or submissive). In Study 2 participants interacted in pairs, and complementarity referred to the match between one participant's interpersonal goals and the other's overt behavior.
Participants whose goals were complemented by their partners' behavior were more satisfied with
the interaction than those whose goals were not. In both studies satisfied participants perceived their
partners as similar to themselves.
Over the years a number of interpersonal theories have
emerged to describe the relationship between two people in an
interaction (e.g., Benjamin, 1974; Homey, 1945; Kiesler, 1983,
1996; Strong & Hills, 1986; Sullivan, 1953; J. S. Wiggins,
1982). Although these theories differ in important ways, most
of them contain the same two fundamental propositions. One
of these propositions is that interpersonal behaviors can be described along two dimensions. The first dimension, affiliation,
ranges from friendliness to hostility; the second, control, ranges
from dominance to submission. Interpersonal behaviors reflect
a combination of these two underlying dimensions. ''Criticizing," for example, lies in the quadrant reflecting hostile dominance, * 'advice-giving" in the quadrant reflecting friendly dominance, "deferring" in the quadrant reflecting friendly submission, and "sulking" in the quadrant reflecting hostile submission. In addition to describing interpersonal behaviors
(Kiesler, 1983; Strong & Hills, 1986), the dimensions of affiliation and control have been used to describe interpersonal problems (Horowitz, 1979), interpersonal traits (Conte & Plutchik,
1981; J. S. Wiggins, 1979), and interpersonal relations (Wish,
Deutsch, & Kaplan, 1976).
The second proposition is that people influence each other's
The preparation of this article was supported by the John D. and
Catherine T. MacArthur Foundation Program on Conscious and Unconscious Mental Processes. It was also supported by Kaiser Foundation
Research Institute Grant 114-9724. We are grateful to David Bergloff,
Ki Chan, Grace Chen, Vincent Ho, and Aie Lopez for their help as
research assistants and to Martin Ford, Edward Haertel, Youngme Moon,
Clifford Nass, Susan Nolen-Hoeksema, and Byron Reeves for their comments on an earlier version of this article.
Correspondence concerning this article should be addressed to D.
Christopher Dryer, who is now at Almaden Research Center, IBM Corporation, 650 Harry Road, San Jose, California 96120, or Leonard M.
Horowitz, Department of Psychology, Jordan Hall, Building 420, Stanford University, Stanford, California 94305-2130.
592
behavior in predictable ways as they interact (Carson, 1969;
Kiesler, 1983; Leary, 1957; Sullivan, 1953). This proposition,
called the principle of complementarity, states that each interpersonal behavior invites a particular class of responses. The behavior and the response it invites are said to be complementary.
Complementary behaviors are similar along the affiliation dimension (friendliness invites friendliness, hostility invites hostility) and reciprocal along the control dimension (dominance
invites submission, submission invites dominance). Therefore,
according to this postulate, advice-giving invites deference, and
sulking invites scolding.
Complementarity in Interactions
Typically, researchers have tested the principle of complementarity by determining whether interpersonal behaviors do, in
fact, complement one another in interactions. Strong et al.
(1988), for example, had participants work with a partner on
a problem-solving task. Raters coded all interpersonal behaviors
and all the reactions to behaviors along the interpersonal circumplex. These researchers hypothesized that "interpersonal actions
tend (with a probability significantly greater than chance)
to . . . evoke from an interactant complementary responses"
(Kiesler, 1983, p. 200). The wording of this hypothesis suggests
a mechanistic view of interpersonal behavior: One behavior
as stimulus evokes another as a reflexive response. However,
interpersonal behaviors vary in ways that violate this mechanistic interpretation. For example. Strong et al. (1988) found that
certain behaviors (e.g., hostile submissiveness) generally elicited noncomplementary responses, though other behaviors
(e.g., friendly dominance) did generally elicit complementary
responses.
Findings such as these indicate that interpersonal behaviors
need not be complementary. Reviews of the empirical tests of
complementarity (e.g., Bluhm, Widiger, & Miele, 1990; Ortbrd,
1986) have suggested that interpersonal behaviors vary considerably in ways not accounted for by complementarity alone.
INTERPERSONAL COMPLEMENTARITY VERSUS SIMILARITY
Theorists have argued that individual differences in interpersonal style (Bluhm et al., 1990) or in interpersonal goals (Horowitz et al., 1991; Orford, 1986) partially determine behaviors.
From this perspective, interpersonal behaviors invite, rather than
elicit or evoke, complementary responses (Horowitz, Dryer, &
Krasnoperova, 1997). People can and do refuse such invitations.
For example, people with a dominant interpersonal style may
refuse an invitation to be submissive, and people with a submissive interpersonal style may refuse an invitation to be dominant.
The noncomplementarity that occurs when one partner refuses to provide the invited response may have distinctive consequences. Noncomplementary interactions may (a) impede joint
productivity (Estroff & Nowicki, 1992; Kiesler, 1983; Sullivan,
1953) and (b) result in dissatisfaction with the interaction (Horowitz et al., 1991; Sullivan, 1953). Recent studies have examined these consequences. For example, Estroff and Nowicki
(1992) found that complementary pairs of participants together
performed better on a jigsaw puzzle task than did noncomplementary pairs. Similarly, Nowicki and Manheim (1991) found
that women in complementary pairs liked each other more after
interacting for 75 min than did women in noncomplementary
pairs. These results suggest that a complementarity between two
partners enhances their attractiveness to each other.
Similarity in Interactions
A contrasting view, the similarity-attraction hypothesis
(Barry, 1970; Blankenship, Hnat, Hess, & Brown, 1984; Duck,
1973; Duck & Craig, 1978; Jellison & Zeisset, 1969; Novak &
Lerner, 1968), states that people are attracted to others who
exhibit similar (rather than complementary) personality characteristics. As Moon (1996) noted, Type A and Type B individuals
prefer dating partners with the same personality type (Morell,
Twillman, & Sullaway, 1989), children prefer others with a
similar sense of humor (Hymel & Woody, 1991), college students prefer roommates with similar personality traits (Carli,
Ganley, & Pierce-Otay, 1991; Deutsch, Sullivan, Sage, & Basile,
1991), and people like strangers with personality characteristics
similar to their own (Byrne & Griffin, 1969; Griffitt, 1966,
1969; Hendrick & Page, 1970; Hodges & Byrne, 1972; Reagor &
Clore, 1970). Likewise, husbands and wives with similar personality characteristics report greater marital satisfaction (e.g.,
Antill, 1983; Barry, 1970; Blazer, 1963; Bruch & Gilligan, 1980;
Eysenck & Wakefield, 1981; Murstein, 1961; Richard, Wakefield, & Lewak, 1990; Tharp, 1963; Wiggins, Moody, & Lederer,
1983). The similarity hypothesis has also been supported when
similarity is defined as shared demographic characteristics,
shared physical characteristics, or shared attitudes (see Moon,
1996, for a review of this literature). The principal theory that
explains the similarity-attraction hypothesis states that a similarity between Person A and Person B is gratifying because each
person validates, enhances, or reinforces the self-concept of the
(similar) other (e.g., Byrne, 1971; Clore & Byrne, 1974).
The principles of similarity and complementarity seem to
contradict one another in some respects but not others. The two
hypotheses agree that similarity with respect to affiliation is
attractive. That is, the principle of complementarity states that
complementary behaviors are similar with respect to affiliation
593
(though reciprocal with respect to dominance). Thus, both
hypotheses imply that friendly or nurturant people should prefer
friendly or nurturant partners. Rirthermore, the two principles
are also compatible in another sense. A man who is unassertive
may dislike his own unassertiveness and prefer partners who
enable him to be more assertive. By the principle of complementarity, an unassertive partner would invite the man to behave
assertively, thereby gratifying his own goals. Thus, the unassertive man and his unassertive partner are similar to each other,
but the mechanism that explains the man's goal satisfaction is
complementarity (Horowitz et al., 1997). This principle would
explain why depressed people prefer other depressed partners
to nondepressed partners (Locke & Horowitz, 1990). Similar
issues have been discussed in greater detail by Horowitz et al.
(1997), and they will be examined more closely in Study 2 of
the present article.
In the present studies we examined whether complementarity
or similarity (with respect to dominance) influences people's
satisfaction with interactions. In order to follow the lead of
earlier researchers in maintaining experimental simplicity (e.g.,
Strong et al., 1988), we limited our observations to dyads of
female students. In Study 1 we observed dominant or submissive
participants interacting with a confederate who enacted a dominant or submissive role. In Study 2 we observed dyads of naive
participants whose interpersonal goals and behaviors were assessed. In both studies, participants reported their satisfaction
with the interaction and their perceptions of their partners. These
studies provided a test of the competing hypotheses by determining whether similarity or complementarity (a kind of dissimilarity with respect to dominance) facilitates interpersonal satisfaction and productivity.
Study 1
Method
Participants
Undergraduate students enrolled in an introductory psychology class
at Stanford University completed a measure of interpersonal dominance
as part of a packet of questionnaires administered to the entire class.
Forty-two female students who had completed the questionnaire packet
participated in this study in partial fulfillment of a class requirement.
Procedure
Overview. All participants were contacted by telephone and scheduled for a laboratory session. When each participant arrived, she met
the experimenter and another female undergraduate (a confederate), who
was to be the participant's partner. The experimenter explained that the
study concerned "interpersonal problem solving and people's satisfaction with interactions.'' After a brief introduction to die study, die experimenter led the participant and her partner to separate rooms. The experimenter asked the participant to read through written instructions for the
procedure.
After the participant had completed this task, the experimenter led
both the participant and her partner to a different room. For their discussion of interpersonal problem solving, each person was presented with
a list of common interpersonal topics from which they were each asked
to select two. Each partner received a different list. The participant's
list of topics included: "It's hard for me to say 'no' to other people"
594
DRYER AND HOROWITZ
and "It's hard for me to stay out of other people's business." These
topics were selected from the Inventory of Interpersonal Problems (Horowitz, Rosenberg, Baer, Ureno, & Villasenor, 1988) such that one topic
was selected from each octant of the interpersonal circumplex (Horowitz, Rosenberg, & Bartholomew, 1993).
The experimenter instructed the partners to draw slips of paper from
a box to determine who was to be the first speaker. This drawing was
contrived such that the confederate was always selected to be the first
speaker. The experimental session consisted of two interactions in which
the confederate presented her two topics for discussion; these are designated Interaction 1 and Interaction 2. Each interaction consisted of four
parts. In Interaction 1 the confederate began by speaking about her first
topic; then the participant reacted; then the confederate responded to the
participant's reaction (the experimental manipulation); and finally, the
participant had an opportunity to react again. Then the entire procedure
was repeated with the second topic to form Interaction 2.
Details of the interaction. The confederate's first topic was "It's
hard for me to feel close to another person." Her script was:
and constructed six hypothetical problem-solving monologues around
each problem. These monologues were typewritten and presented to
students in an introductory psychology class at Stanford University. The
students were asked to use a 5-point scale to rate the speaker of each
monologue along the dimensions of control (' 'How submissive or dominant was the speaker?") and affiliation ("How cold or friendly was the
speaker?"). Our goal was to identify the two topics that did not a priori
make the speaker seem dominant or submissive. The topic monologues
we selected were the ones that best met this criterion.
Then we created four complete scripts for each topic monologue by
adding one of four possible responses that a speaker might make to any
hypothetical listener reaction. The four complete scripts were presented
to another introductory psychology class. On a 5-point scale, the students
rated each of the four responses along the two interpersonal dimensions.
From these ratings we identified the response for each script that made
the speaker seem dominant (but neutral along the affiliation dimension)
and the response that made the speaker seem submissive (but neutral
along the affiliation dimension). We called the two selected scripts,
respectively, the dominant role script and the submissive role script.
Okay. I'm going to pick "it's hard for me to feel close to another
person." (pause) Well, f've been seeing this guy for a couple of
months now, and I just feel like he's more into it than I am. 1 want
to feel close to him, but I just can't create emotions that aren't
there. It's not anything about htm because he really is a great
boyfriend—I mean I like him a lot and he's good looking and all
that—it's just that I know I'm not feeling the same thing he is. I
don't really know what to do about it because I've had this problem
before. It's not that I don't like him because I really do and part
of me really like wants to be with him, but, I don't know; I just
feel pressured to feel something I don't. I don't know what to say
because I want to feel close but I can't.
Four confederates were trained over a 2-week period to enact each
of the two scripts. Each confederate interacted with approximately equal
numbers of participants, and the participants were assigned randomly
to one of the two conditions (dominant vs. submissive confederate role).
None of the participants had met the confederate previously or expressed
any suspicion about the confederate after the experimental session.
After the confederate recited this script, the participant had an opportunity to say whatever she wished. Then the partners were told to rate
their satisfaction with the interaction so far. We called this rating the
before rating because it occurred before the experimental manipulation.
Then the confederate responded with an apparently impromptu remark,
which served as the experimental manipulation. In the submissive rote
condition, she said: "Yeah, what you say makes sense. I don't know
why it is, but I feel that I just haven't been able to solve this problem
on my own. 1 guess I really do need help." In the dominant role condition, she said: "Yeah, this is one problem that I'll just have to solve on
my own. I guess nobody can solve personal problems for you; you have
to figure them out for yourself." The confederate's behavior was identical in the two conditions except for the above responses.
The participant then had an opportunity to react again, and that concluded Interaction 1. Then the partners rated their satisfaction with the
interaction again. This rating occurred after the experimental manipulation, so we called it the after rating. The entire procedure was repeated
again to form Interaction 2. This time the confederate's topic was "LTt's
hard for me to forgive another person after I've been angry." Her script
for this topic was similar to the one presented above,
After the interactions were over, the experimenter led the participant
and her partner to separate rooms, and the participant rated the interpersonal style of her partner on the Bern Sex Role Inventory (BSRI; Bern,
1974; described below). Then the experimenter debriefed the participant,
explaining the experimental procedure and determining whether she had
any suspicions about her partner. Finally, the experimenter brought the
participant and the confederate together again, asked the participant if
she had any further questions, and thanked her for her involvement.
Construction of the Confederates' Scripts
To construct the confederates' scripts, we selected problem statements
from the Inventory of Interpersonal Problems (Horowitz et al., 1988)
Measures
The participants' interpersonal styles were assessed with the BSRI.
The Masculinity subscale of the BSRI correlates almost perfectly with
other measures of interpersonal dominance (J. S. Wiggins &. Broughton,
1985), and it has high internal consistency (for this sample, Cronbach's
a = .88). A median split of this measure divided participants into a
dominant style group and a submissive style group. We used a modified
version of this questionnaire to assess participants' perceptions of the
confederate's interpersonal style.
To assess the participants' satisfaction with the interaction, a series
of buttons was built into the table in front of the interactants. The buttons
were arranged so that each person could see her own buttons but not
those of her partner. The participants indicated their satisfaction as well
as their frustration by pressing buttons with the appropriate labels (not
at all satisfied to extremely satisfied for one set, not at all frustrated to
extremely frustrated for the other set). Because the two were significantly correlated (r — —.64, p < .01), the frustration ratings (F) were
combined with the satisfaction ratings (S) to form a 2-item ( S - F )
measure of satisfaction.
Results
Satisfaction With the Interactions
Satisfaction with the first interaction. In Interaction 1, participants rated their satisfaction before and after the experimental manipulation. The mean satisfaction ratings are shown in
Table 1. Before the experimental manipulation, all participants
were interacting with interpersonally neutral confederates, so
no initial differences were expected for the before rating. The
data showed no significant differences among the four groups.
A 2 (confederate's role) x 2 (participant's style) analysis of
variance (ANOVA) revealed no main effects and no interaction
effect, all F{ 1, 38)s < 1.5, all ps > ,30.
After the manipulation, participants made another satisfaction
595
INTERPERSONAL COMPLEMENTARITY VERSUS SIMILARITY
Table 1
Mean Ratings of Satisfaction With the First Interaction
Judgments of the Confederate's Interaction Style
Confederate's role
Participant's style
Dominant
Submissive
Before experimental manipulation
Dominant
Submissive
2.17"
2.18b
1.73"
2.5(T
After experimental manipulation
Dominant
Submissive
1.83s
2.91 b
declined or stayed the same in the other two partnerships (the
mean difference scores were 0.00 and - 1 . 2 5 ) .
2.73b
1.75C
Note. Higher ratings indicate more satisfaction.
*n = 12. b « = 11. cn = 8.
rating. The means for the after ratings are also shown in Table
1. Here, the effect of the experimental treatment is clear. A 2 X
2 analysis of covariance (ANCOVA) with the before rating as
a covariate revealed a significant interaction effect, F( 1, 38) =
9.45, p < .005 (all other Fs < 1). Dominant participants who
interacted with submissive confederates and submissive participants who interacted with dominant confederates were more
satisfied with the interaction than were participants in the other
two conditions. A planned orthogonal contrast revealed a significant difference between the complementary and the noncomplementary groups, r(41) = 3.31, p < .005. Lt is important to
note that the satisfaction ratings improved for the two complementary partnerships (the mean difference scores were 1.00 and
0.73), whereas the satisfaction ratings declined in the other
two partnerships (the mean difference scores were —0.34 and
-0.75).
Satisfaction with the second interaction. After the participants completed the first interaction, the confederate began the
second interaction by selecting a different topic. As with the
first interaction, the confederate's script elaborated the topic.
The confederate's script was initially interpersonally neutral
(i.e., neither dominant nor submissive). The mean satisfaction
ratings for the second interaction are shown in Table 2. We used
a 2 (confederate's role) X 2 (participant's style) ANOVA to
analyze the before ratings. As with Interaction 1, this analysis
revealed no main effects and no interaction effect. After the
manipulation, participants made another satisfaction rating. The
means for the after ratings are also shown in Table 2. As with
Interaction 1, the effect of complementarity was again evident.
A 2 X 2 ANCOVA with the before rating as a covariate revealed
a significant interaction effect, F( 1,38) = 8.26,/? < .01, Dominant participants who interacted with submissive confederates
and submissive participants who interacted with dominant confederates were more satisfied with the interaction than were
participants in the other two conditions. A planned orthogonal
contrast revealed a significant difference between the complementary and the noncomplementary groups, f(41) = 2.75, p <
.01. lt is important to note that the satisfaction ratings improved
for the two complementary partnerships (the mean difference
scores were 0.28 and 0.54), whereas the satisfaction ratings
The above results show that a participant whose partner's
reactions complemented her own interaction style was more
satisfied with the interaction than a noncomplemented partner.
However, we were curious to learn whether such a participant
judged her partner to be (a) complementary to herself (the
complementarity hypothesis) or (b) similar to herself (the similarity hypothesis). Did satisfied dominant participants, for example, judge their partners to be submissive on the BSRI (the
complementarity hypothesis) or dominant like themselves (the
similarity hypothesis)? Therefore, we divided the dominant and
submissive participant groups into more satisfied and less satisfied subgroups, using a median split of their mean rating of
satisfaction. Of the 23 dominant participants, 11 were more
satisfied, and 12 were less satisfied; of the 19 submissive participants, 8 were more satisfied, and 11 were less satisfied.
A 2 (participant's dominance) X 2 (satisfaction) ANOVA
revealed only a significant interaction effect, F(\, 38) - 5.20,
p < .05. Satisfied submissive participants judged their partners
to be submissive (like themselves); their mean rating of dominance was -1.39. Satisfied dominant participants judged their
partners to be dominant (like themselves); their mean rating
was 3.93. For the dissatisfied groups, the corresponding mean
ratings of their partners' dominance were 5.07 for the submissive
participants and —0.91 for the dominant participants. This finding supports the similarity hypothesis: Satisfied participants,
regardless of their partners' behavior, judged their partners to be
similar to themselves. Correspondingly, dissatisfied participants
judged their partners to be dissimilar to themselves. The finding
resembles other results (e.g., Kenny, 1994) that showed that
people who like each other judge their partners to be like themselves. It also suggests that the complementarity that produced
the participants' satisfaction is not evident in the participants'
descriptions of their partners.
Discussion
In this study the participants were selected according to their
interpersonal styles, as assessed by the BSRI. Participants who
Table 2
Mean Ratings of Satisfaction With the Second Interaction
Confederate's role
Participant's style
Dominant
Submissive
Before experimental manipulation
Dominant
Submissive
2.50a
1.82"
2.45b
1.62C
After experimental manipulation
Dominant
Submissive
1.25*
2.36b
Note. Higher ratings indicate more satisfaction.
" „ = 12. b n = 11. cn = 8.
2.73b
1.62C
596
DRYER AND HOROWITZ
described themselves as dominant were more satisfied when
they interacted with a partner whose behavior was explicitly
submissive, and submissive participants were more satisfied with
a dominant partner. The results of this study indicate that complementarity with respect to dominance does influence people's
satisfaction with an interpersonal interaction.
Despite this influence of objective complementarity, however,
the participants did not judge complementary partners to be dissimilar to themselves in dominance. Instead, satisfied participants
judged their partners to be similar to themselves, and dissatisfied
participants judged their partners to be dissimilar. We therefore
need to differentiate between objective complementarity (which
leads to satisfaction) and perceived similarity (which seems to
follow from complementarity). Perhaps people do not realize the
basis for their satisfaction, but when they are satisfied, they regard
their partners as being like themselves. In some contexts the rule
may be valid that similarity leads to liking, but in this context,
liking apparently leads to perceived similarity. We tested (and
replicated) this phenomenon in Study 2.
Study 1 had two limitations. First, the participant's partner
was a confederate whose reaction to the participant was unambiguous (explicitly dominant or explicitly submissive), and we
do not know whether we can generalize from reactions in this
contrived experimental situation to people's reactions in a more
natural conversation. Tn Study 2 we therefore observed participants interacting more casually with each other. Second, the
participants in Study 1 were classified as dominant or submissive on the basis of their self-descriptions on a personality inventory. As noted earlier, however, personality traits are ambiguous
with respect to people's goals. When people describe themselves
as dominant, for example, they may be describing their goal
behaviors, their actual behaviors, or some other aspect of their
social behavior. Therefore, in Study 2 we specifically assessed
people's interpersonal goals (wishes) as well as their actual
overt behavior.
measures as a guide (Horowitz et al., 1988; Kiesler, 1983; Strong &
Hills, 1986). Items were considered acceptable only if they reflected an
unambiguous goal and were endorsed by many people on pilot versions.
Fifty-one final items were presented as a questionnaire that asked respondents to "imagine that you are working with another person on a task that
is important to you." Each item had the stem: "It would be important to
me to . . ." Respondents indicated their agreement with each item by
circling one of five numbers from 0 to 4, with 0 indicating no, definitely
not; 2 indicating hard to say; it depends; and 4 indicating yes, definitely.
The format of the questionnaire and the wording of the items were based
on the Assessment of Personal Goals (APG; Ford & Nichols, 1991).
Study 2
DGS = PA + .7BC - .7FG - HI - .7JK + .7NO.
The first purpose of Study 2 was to test the hypothesis that
satisfaction depends on the match between a participant's interpersonal goals and his or her partner's behavior: People who
want to dominate their partner but observe cues that their partner
wishes to dominate them know that their goal to dominate will
not be fulfilled. However, no measure currently exists that assesses interpersonal goals. Therefore, we needed to construct a
measure of interpersonal goals to test the hypothesis. Constructing that measure was a second purpose of the study.
Study 2 also extended the findings of Study 1 in two general
ways. First, the participants interacted naturalistically with another participant (rather than a confederate); second, they
worked together on a task that allowed the quality of their joint
performance to be evaluated objectively. These two extensions
helped broaden the generalizability and scope of Study 1.
Method
Constructing the Interpersonal Goals Inventory (IGI)
Selecting items. To develop the IGI, we first identified a sample of
items that spanned the interpersonal domain, using other interpersonal
Participants and procedure. The questionnaire was administered (o
205 Stanford undergraduate students as part of a battery of questionnaires. Respondents partially fulfilled an introductory psychology requirement through their participation.
Construction of subscales. We followed the procedure used by other
investigators (e.g., Alden, Wiggins, & Pincus, 1990) to generate eight
subscales that met the criteria of a circumplex. We ipsatized and standardized participants' responses to the items and then subjected the item
intercorrelations to a principal-components analysis. As usual in such
analyses, the first principal component reflected dominance (in this case,
dominance goals), and the second principal component reflected affiliation (in this case, affiliation goals). Next, we plotted each of the 51
items in a two-dimensional space, using each item's loadings on each
of the first two principal components. A two-dimensional interpersonal
space is customarily divided into eight equal sectors (octants) with
theoretical midpoints at 90°, 135°, 180°, 225°, 270°, 315°, 0°, and 45°.
These octants are designated PA, BC, DE, FG, HI, JK, LM, and NO,
respectively.1 The four items having the highest communality within
each 45° octant formed a subscale. The items in each of the eight
subscales are listed in the Appendix. We calculated subscale scores for
each participant by adding together the participant's responses to the
four items in each of the eight subscales.
Then we computed a composite score for the dominance goals by
weighting each subscale score by its theoretical coordinate along the
dominance dimension. Following conventional procedures (Gurtman,
1991), we defined a composite dominance goals score (DGS) as the
sum of the weighted subscale scores, where each subscale score is
weighted by the cosine of its angular location on the interpersonal circle.
A high DGS indicates that behaving dominantly is important to the
respondent, and a low DGS indicates that behaving submissively is
important to the respondent. For die experimental sample described
below, the DGS ranged from -3.18 to 2.22, with a mean of -0.02 and
a standard deviation of 1.34. The test-retest reliability across two testings (a 2- to 5-week interval) was .79. For a complete description of
the construction and validation of the IGI, see Dryer (1993).
Participants
The IGI was administered to Stanford introductory psychology students as pare of a battery of questionnaire materials. The students completed the battery in partial fulfillment of a course requirement. We
identified 48 women whose goals were clearly dominant (i.e., scores
1
Earlier investigators, such as Leary (1957), divided the interpersonal
space into 16 sectors designated A, B, C . . . P. However, subsequent
research has shown that a division into eighths (octants), rather than
sixteenths, is preferable. Therefore, adjacent sixteenths have been combined to form octants: the sixteenths that had been labeled B and C have
become the octant BC; D and E, the octant DE; and so on.
INTERPERSONAL COMPLEMENTARITY VERSUS SIMILARITY
that were at least 0.5 SD above the IGI mean score for dominance goals)
or clearly submissive (i.e., scores that were at least 0.5 SD below the
IGI mean score for dominance goals). We randomly assigned the women
to 24 pairs. Two pairs were excluded from analysis because they had
known each other previously, leaving a total of 44 participants.
Experimental Procedure
Research assistants invited the women to participate in a study on
"problem solving," telling them that they would "work together with
another undergraduate on a couple of tasks." When the two partners
arrived at the laboratory, the experimenter handed them consent forms
and briefly described the study. The procedure consisted of four main
parts: (1) preinteraction questionnaires, (2) problem solving interaction,
(3) postinteraction questionnaires, and (4) debriefing,
Preinteraction questionnaires. Before their interaction, the women
each completed a battery of questionnaires. The battery consisted of a
second IGI, a set of personality rating scales (Malle & Horowitz, 1995),
and the Desert Survival Problem (DSP; Lafferty & Eady, 1974). The
personality rating scales consisted of a list of 32 trait adjectives, describing the negative and positive poles of four of the Big Five personality
dimensions (John, 1990; McCrae & Costa, 1989): Extraversion, Agreeableness, Conscientiousness, and Neuroticism. The participants indicated
how self-descriptive they thought each item was (from 1 = not at all
to 9 = extremely). The final questionnaire in the preinteraction battery
was the DSP, a problem-solving task in which the participants were to
imagine that they were copilots of an airplane that had crash-landed in
the middle of a desert. Several items (e.g., a cosmetic mirror, a map,
two raincoats) could be rescued from the plane before it was consumed
in flames, and die participants' task was to rank order these 10 items
such that the item most important to survival was ranked first, the second
most important item was ranked second, and so on.
Problem-solving interaction. After the participants completed the
preinteraction questionnaires, the experimenter led them into another
laboratory room, where they worked on two tasks together. First, as a
practice task, the two participants had to rank order five fruits (kiwi,
banana, apple, lemon, and watermelon) as to how well each one exemplified the category "fruit." Then the experimenter instructed the participants to work on the DSP together. They were informed that an optimal
solution to the problem does exist (Lafferty & Eady, 1974) and that
any partners who produced the optimal solution would receive a $100
bonus. (None did.) The experimenter then left them alone to work on
the DSP together. A typical interaction included a range of interpersonal
behaviors. Typically, both participants suggested items that they considered very important or very unimportant and offered reasons for their
opinions. In general, they quickly identified differences in their opinions,
discussed their views, and reached some agreement. Sometimes arguments were challenged, other times they were not. The task required
some amount of cooperation, and every dyad managed to generate a
joint solution to the problem.
During each interaction, a prerecorded voice periodically instructed
the participants to rate their satisfaction with the interaction. To indicate
their ratings, participants pressed buttons built into the table in front of
them. There were two sets of five buttons for each satisfaction rating.
The buttons in one set had labels indicating the degree of satisfaction,
ranging from (1) not at all satisfied to (5) extremely satisfied. Those
in the other set had labels indicating the degree of frustration, ranging
from (1) not at all frustrated to (5) extremely frustrated. Each participant could see her own but not her partner's buttons. The participants
had 3 min to solve the practice problem and 16 min to solve the DSP.
The prerecorded voice instructed them to rate their satisfaction at the
beginning of each problem, and then after every 90 s for the practice
problem and after every 4 min for the DSP. The voice also indicated
597
how much time was remaining. The participants were aware that their
interactions were being videotaped.
Three research assistants, who were aware neither of the hypotheses
nor of the participants' self-rated goals, later viewed the videotapes and
rated the dominance of the participants* behaviors. These raters first
studied theoretical descriptions of interpersonal behavior prepared by
Kiesler (1983) and by Strong and Hills (1986). They were also shown
the videotaped behavior of participants in an earlier study, selected
because of their extreme dominance or submissiveness. The raters were
also instructed to consider certain specific behaviors as evidence of
dominance (advice-giving, criticizing, asserting one's own beliefs, expressing self-confidence). Similarly, they were to consider other specific
behaviors as evidence of submissiveness (self-criticizing, deferring, focusing on partner beliefs, expressing self-doubt). The raters then made
practice ratings of the behavioral dominance of pilot participants, until
they reached an acceptable level of agreement. The raters used verbal
and nonverbal behavior as evidence for their ratings, assigning a rating
of "dominant" or "submissive" to each participant.
Postinteraction questionnaires. After the two problem-solving interactions, the participants worked independently on two final sets of questionnaires. They indicated their perceptions of their partners' goals and
personality by completing modified versions of the IGI and the personality rating scales. The IGI-Partner version contained items in the form
"It would be important to my partner to . . ." The personality rating
scales required the participants to indicate how descriptive each trait
adjective was of their partner. Additionally, they completed a 5-item
questionnaire that assessed their global satisfaction with the interaction.
Debriefing. After the participants completed the postinteraction
questionnaires, the experimenter described the rationale for the study.
When their questions had been answered to their satisfaction, the experimenter asked them not to reveal the nature of the tasks to other students
and thanked them for their help.
Results
Description of Sample and Measures
Interpersonal goals. The participants in this study had been
selected to have high or low dominance goals as assessed by
the IGI. The z scores for dominance goals ranged from —3.51
to -0.48 for the submissive goal group, and from 0.59 to 2.22
for the dominant goal group.
Observer ratings of behavior. The three coders showed considerable agreement in their ratings (Cronbach's a = .77). To
increase the reliability of the measure (Horowitz, Inouye, &
Siegelman, 1979), we combined the three ratings of dominance
to form a single measure. Participants were classified as ' 'behaviorally dominant" if at least two coders rated their behavior as
dominant. They were classified as "behaviorally submissive"
if at least two coders rated their behavior as submissive.
Measures of satisfaction. Participants periodically rated
both their satisfaction and their frustration with the interaction,
and their ratings on these two measures were negatively correlated; the mean r across trials was —.60. To increase the reliability of the measure, we combined the satisfaction and frustration
ratings into a single score; the sign of the frustration ratings
was reversed, and the two were summed. This measure is called
the satisfaction rating. In addition, the participants' responses
to each of the five items of the satisfaction questionnaire were
positively correlated with their satisfaction ratings; the mean r
across both trials and items was .27. To obtain a single, more
598
DRYER AND HOROWITZ
reliable measure of satisfaction, we combined responses to the
questionnaire with the satisfaction ratings. Scores on each measure were first standardized and then added together. Cronbach's
alpha for the overall measure of satisfaction (hereafter called
composite satisfaction) was .84. This way of aggregating measures follows the convention in recent research (e.g., Locke &
Horowitz, 1990; Moon, 1996). The increase in internal consistency that results from aggregating measures in this way is
known to increase the validity of the resulting measure (Horowitz et al., 1979); it also avoids treating nonindependent outcome measures as though they were independent.
Impact of Complementarity on Satisfaction
Composite satisfaction. We then examined the impact of
interpersonal complementarity on the participants' composite
satisfaction scores. First we determined whether the two partners' composite satisfaction scores were interrelated. We computed an intraclass correlation coefficient (McNemar, 1962).
The value was .27, which was not significant, f (21, 22) =
1.73, p > .10. For this reason, we performed two kinds of
analyses. The first treated the partners' composite satisfaction
scores as though they were independent measures (Kraemer &
Jacklin, 1979). The second (also following Kraemer & Jacklin,
1979) divided the partners into two independent samples
(thereby avoiding an assumption of independence). As described below, the second analysis confirmed the findings of the
first.
We prepared a 2 X 2 matrix, classifying each participant
according to her own interaction goals (dominant or submissive)
and her partner's actual behavior (dominant or submissive).
The mean composite satisfaction score for each cell is shown
in Table 3. Then we performed a 2 (participant's dominance
goals) x 2 (partner's behavior) ANOVA on the composite satisfaction scores. Neither main effect was significant, but the interaction was; F ( 1 , 40) = 7.585, p < .01.
To examine this interaction further, a planned orthogonal contrast compared goal-complementary participants with goal-noncomplementary participants. That is, it compared (a) participants with dominant goals interacting with submissive partners
and participants with submissive goals interacting with dominant partners with (b) participants with dominant goals interacting with dominant partners and participants with submissive
goals interacting with submissive partners. The mean satisfaction for the goal-complementary participants was significantly
Table 3
Mean Composite Satisfaction Ratings
Partner's behavior
Participant's
interpersonal goals
Dominant
Submissive
Dominant
a
-.39
.82C
Note. Higher ratings indicate more satisfaction.
a
n = 12. b n = U. cn = 1. dn = 14.
Submissive
.31 b
-.83 d
higher than that for the goal-non complementary participants,
r(40) = 2.75, p < .01.
Then, to circumvent the assumption of independence, we created two subsamples following the procedure recommended by
Kraemer and Jacklin (1979). One partner of each dyad was
selected at random to form Subsample 1; the remaining partners
formed Subsample 2. We then analyzed the data from each
subsample separately, using a one-tailed test. The results and
conclusion for each subsample were similar to those described
above. The goal-complementary participants were more satisfied
with the interaction than were the goal-noncomplementary participants; for Subsample 1, r(18) = 1.90, p < .05, and for Subsample 2, r(18) = 1.83,/? < .05.
These results demonstrate the importance of the match between the goals of one partner and the behavior of the other in
determining their satisfaction. When they match, the partners
are more satisfied with the interaction. Perhaps satisfaction
could also be predicted just as well from other interpretations
of complementarity—that is, from the complementarity between the two partners' behaviors or from the complementarity
between their goals. To test these alternate interpretations, we
first examined the complementarity of the participants' goals.
We prepared a 2 x 2 matrix, classifying each participant according to her own interaction goals (dominant or submissive)
and her partner's interaction goals (dominant or submissive).
The mean composite satisfaction scores for the dominant-dominant, dominant-submissive, submissive-dominant, and submissive-submissive conditions, respectively, were 0.26 (n =
14), -0.55 (n = 9), -0.52 (n = 9), and -0.11 (n - 12). We
performed a 2 (participant's dominance goals) X 2 (partner's
dominance goals) ANO\A on the composite satisfaction scores.
All effects were nonsignificant (Fs < 1.9, ps > .15). The
match between the participants' goals did not determine their
satisfaction with the interaction.
Then we examined the complementarity of the participants'
behaviors. We prepared a 2 X 2 matrix, classifying each participant according to her own behavior (dominant or submissive)
and her partner's behavior (dominant or submissive). The mean
composite satisfaction scores for the dominant-dominant, dominant-submissive, submissive-dominant, and submissive-submissive conditions, respectively, were -0.69 (n - 6), -0.04 («
= 13), 0.43 (n = 13), and -0.67 (n - 12). Neither main
effect was significant, and the interaction was only marginally
significant, F ( l , 40) = 3.89, p < .10. Dominant participants
were (marginally) more satisfied when their partners behaved
submissively, and submissive participants were (marginally)
more satisfied when their partners behaved dominantly. Thus,
the clearest predictor of satisfaction was the match between a
participant's goals and her partner's behavior.
We were therefore curious to examine the relationship between the participants' self-reported dominance goals and actual
behavior (as rated by objective judges). We prepared a 2 x 2
matrix, classifying each participant according to her self-reported goals (high vs. low dominance goals) and her own behavior (high vs. low dominance). Of the 21 participants with
low dominance goals, 13 exhibited low dominance behavior,
and 8 exhibited high dominance behavior. Of the 23 participants
with high dominance goals, 12 exhibited low dominance behav-
INTERPERSONAL COMPLEMENTARITY VERSUS SIMILARITY
ior; and 11 exhibited high dominance behavior. The relationship
between the two variables was not statistically significant,
X2(l,N = 44) = 0.42.
Satisfaction across time. We then examined the data to determine whether the effect of the goals-behavior complementarity on satisfaction changed in the course of the interaction. The
participants had rated their satisfaction at the beginning of the
interaction and at three more points during the interaction. The
partners' corresponding ratings were correlated on each of the
four occasions. The intraclass correlations were, respectively,
.50, .26, .39, and .52 (all ps < .10). Because they were positively correlated, we averaged the two partners' satisfaction ratings on each occasion to create four dyad satisfaction ratings.
Finally, we classified each dyad as 0, 1, or 2 according to its
number of goal-complementary partners—0 indicated that neither partner was goal-complementary; 1, that one partner was
goal-complementary (and the other was not); and 2, that both
partners were goal-complementary. We called these three groups
of dyads fully noncomplementary, mixed, and fully complementary. There were 7,12, and 3 dyads in each of these three groups,
respectively.
To demonstrate that the three groups were initially comparable, we performed a one-way ANOVA on the initial satisfaction
ratings (just after the practice task but before the experimental
task). The ratings of the three groups did not differ significantly,
f ( 2 , 19) < 1. Then we analyzed the satisfaction ratings of the
three groups at the three time periods with a 3 (groups) X 3
(time periods) mixed design ANOVA, which revealed a significant main effect of groups, F(2,19) = 4.06, p < .05. Although
the interaction with time periods did not reach significance, we
performed a separate one-way ANOVA on the data at each time
period. The values of F(2, 19) for the effect of groups at each
of the three time periods, respectively, were 2.01,2.84, and 4.89.
The mean satisfaction ratings (averaged across the three time
periods) were 6.67 for the fully complementary group, 5.80 for
the mixed group, and 3.67 for the fully noncomplementary
group. Thus, after the initial period, the fully complementary
group reported the highest satisfaction, and the fully noncomplementary group reported the lowest. We also compared the group
differences using a planned orthogonal polynomial contrast of
the three levels of the group factor. This contrast indicated that
each of the three groups differed significantly from the other
two. The linear effect was significant, f( 19) = 2.45, p < .05,
and the quadratic effect (i.e., the departure from linearity) also
was significant, *(19) = 2.37, p < .05.
Because the fully complementary group was so small, we
combined those dyads with the mixed dyads and compared the
resulting group with the fully noncomplementary group. The
mean satisfaction ratings for the noncomplementary dyads on
each of the three occasions were 4.29, 3.43, and 3.29; the corresponding means for the complementary group were 5.87, 5.80,
and 6.27. We analyzed the ratings at each of the three successive
time periods with a 2 (groups) X 3 (time periods) mixed design
ANOVA. The main effect of groups was significant, F( 1, 20)
= 7.77, p = .01. We performed a separate one-way ANOVA on
the data at each time period. The value of F(\, 20) for the
effect of groups at each time period was, respectively, 3.90,
5.26, and 9.60.
599
Impact of Complementarity on Task Performance
We also evaluated the quality of the participants' task performance. To do so, we compared the expert solution (Lafferty &
Eady, 1974) with that of (a) the participants' individual solutions and (b) the dyads' joint solutions. We used Pearson rank
order correlation coefficients (p; McNemar, 1962) to assess the
correspondence between the solution of the participants and
that of the expert. We called the correlation between individual
solutions and the expert solution the hefore-interaction score,
one for each member of the dyad. The higher this score, the
higher the quality of that participant's solution. Similarly, we
called the correlation between a dyad's joint solution and the
expert solution the after-interaction score. We then computed
two intraclass correlations: (1) that between the corresponding
before-interaction scores of the two partners and (2) that between the partners' amount of gain from the before- to the afterinteraction scores. Neither intraclass correlation approached significance (both p < .20, p > .10), so the solution scores could
be analyzed as independent measures (Kraemer & Jacklin,
1979).
We analyzed the after-interaction scores with a 3-factor
(groups) ANCOVA, using the before-interaction scores as a
covariate. The results showed two significant effects. First, the
after-interaction scores tended to covary with the before-interaction scores, F ( l , 40) = 8.01, p < .01. Second, the groups
differed in their performance, F(2, 40) - 3.84, p < .05. The
mean improvement on this performance measure was 0.16 for
the fully complementary group, 0.19 for the mixed group, and
0.27 for the fully noncomplementary group. Thus, the least
satisfied participants seemed to show greater gain from their
preinteraction performance on the DSP than those paired with
complementary partners. Apparently, a complementarity between partners, although satisfying, may not necessarily facilitate a dyad's productivity; in fact, noncomplementary partners
may perform better.
Impact of Satisfaction on Perceptions of Partners
In Study 1 the satisfied participants perceived their partners
to be more similar to themselves than did dissatisfied participants. To test whether the same effect occurred in Study 2, we
classified the participants as satisfied or dissatisfied, using a
median split of their satisfaction ratings. We also classified them
as high dominant or low dominant, using a median split of their
dominance goal ratings. These classifications resulted in four
groups: (1) satisfied, dominant goals; (2) satisfied, submissive
goals; (3) dissatisfied, dominant goals; and (4) dissatisfied,
submissive goals. The ns, respectively, were 11, 8, 11, and 14.
For each of the four groups, we computed the participants'
mean rating of their partners' dominance goals and analyzed
these ratings with a 2 (high vs. low dominance goals) x 2 (high
vs. low satisfaction) ANOVA. Among participants who were
satisfied with their interactions, the dominant ones rated their
partners 2.12 in dominance goals, and the submissive participants rated their partners — 1.59. Among participants who were
dissatisfied, the corresponding means were —1.10 (dominant
participants) and 0.56 (submissive participants). The interac-
600
DRYER AND HOROWITZ
tion was significant between dominance goals and level of satisfaction, F(i, 40) = 4.31,/? < .05. A planned orthogonal contrast
revealed that satisfied participants rated their partners to be like
themselves, whereas dissatisfied participants rated their partners
to be different from themselves, r(42) = 2.08, p < .05. This
result suggests that participants used the similarity principle to
rate their partners, as they had done in Study 1.
We performed a similar analysis to compare satisfied and
dissatisfied participants' ratings of their partners' dominance
(the trait). This time, we divided the participants into groups
high or low in self-reported dominance, so as to produce four
new groups: (1) satisfied participants who called themselves
dominant, (2) satisfied participants who called themselves submissive, (3) dissatisfied participants who called themselves
dominant, and (4) dissatisfied participants who called themselves submissive. The ns, respectively, were 11, 8, 11, and 14.
The participants' mean ratings of their partners' dominance for
the four groups, respectively, were: (1) 1.65, (2) —2.56. (3)
—2.66, and (4) 1.49. We also analyzed these ratings with a 2
(trait dominance groups) X 2 (satisfaction groups) ANO\A.
The interaction between trait dominance and level of satisfaction
was significant, F ( l , 4 0 ) - 8.17,/7 < .01. A planned orthogonal
contrast revealed that satisfied participants judged their partners
to be similar to themselves, and dissatisfied participants judged
their partners to be dissimilar to themselves, r(42) = 2.86, p
< .01. These results thus confirmed the corresponding finding
of Study 1.
Because the participants' perception of their partners appeared to be influenced by their own satisfaction, we were curious to examine the relationship between each participant's reported perception and her partner's actual behavior. We prepared
a 2 X 2 matrix, classifying each participant according to her
behavior (high vs. low dominance) and her partner's perception
of her behavior (high vs. low dominance). Of the 25 participants
who exhibited low dominance behavior, 14 were perceived as
low dominance, and 11 were perceived as high dominance. Of
the 19 participants who exhibited high dominance behavior, 8
were perceived as low dominance, and 11 were perceived as
high dominance. The relationship between the two variables
was not statistically significant, x 2 (l> & = 44) = 0.83.
Discussion
In this study we examined the joint influence of a person's
interpersonal goals and the partner's interpersonal behavior on
that person's satisfaction with the interaction. People who had
a goal to dominate were more satisfied when they interacted
with a partner who behaved submissively than with one who
behaved dominantly. Conversely, people who had a goal to submit were more satisfied when they interacted with a partner who
behaved dominantly than with one who behaved submissively.
These results support the claim that people have certain preferred ways of interacting that determine their satisfaction with
an interaction.
Why should people with dominant goals be more satisfied
with submissive partners? We believe that every interpersonal
behavior invites, intentionally or not, a particular reaction from
the partner. A dominant behavior, for example, invites a submis-
sive response, and a submissive behavior invites a dominant
response. The invited response may or may not match the partner's interpersonal goals. If Person B's behavior invites a reaction from Person A that matches A's goal, then A is satisfied;
otherwise, A is frustrated.
Although these results support the principle of complementarity, the participants' descriptions of partners whom they found
satisfying paradoxically supported the principle of similarity.
Although participants with dominant goals were more satisfied
interacting with submissive partners {as judged by objective
observers), those participants described their partners as dominant (like themselves). Thus, in this study, as in Study 1, "liking" led to "perceived similarity." Why is "liking" related to
the perception of "being like"? One possibility (Byrne, 1971;
Schachter, 1959) is that similar people generally share common
beliefs, expectations, and values. When people perceive that
their partner is similar to themselves, they assume that they
and their partner will have "common ground" (Clark, 1992),
"companionate expectations" (Krokoff, Gottman, & Roy,
1988), or "co-orientation" (McLeod & Chaffee, 1973) and,
because of this assumption, they anticipate a smooth interaction:
They assume that they will understand their partner's goals and
that their partner will understand their goals. Perhaps any circumstance that promotes a sense of overlapping goals enhances
a person's satisfaction with a partner. In some cases similarity
(e.g., similar values) implies compatible goals, and in other
cases complementarity implies compatible goals.
General Discussion
In both studies, people preferred to interact with a partner
who was complementary with respect to dominance. Dominant
people preferred to interact with partners who invited them to
be dominant, and submissive people preferred to interact with
partners who invited them to be submissive. This result occurred
(a) when participants interacted with a partner-confederate enacting a scripted role (Study 1) and (b) when they interacted
with another naive participant (Study 2).
Certain results require a reinterpretation of the principle of
complementarity. If we interpret the principle to mean that complementary behaviors necessarily lead to greater satisfaction, it
must be rejected. The dominating behavior of one person together with the submissive behavior of the other is no guarantee
of partner satisfaction. In Study 2, neither a behavior-behavior
complementarity nor a goals-goals complementarity was as
good a predictor of satisfaction as a goals-behavior complementarity. Apparently, behavioral complementarity (e.g., submissive behavior in one person, dominating behavior in the
other) is not, in and of itself, sufficient to guarantee satisfaction;
a person's goals must be considered as well.
Why is satisfaction better predicted when we know the goals
of one partner and the behavior of the other? Person A's goals
are internal events, not directly observable by Person B, su they
can only be inferred from A's behavior. If A's behavior accurately reflects his or her goals, then B would have no difficulty
discerning and satisfying A's goals. But A's behavior may be
ambiguous or even misleading, fbr example, A may be conflicted about his or her wish to dominate and may therefore
INTERPERSONAL COMPLEMENTARITY VERSUS SIMILARITY
camouflage that wish behaviorally. Indeed, A may be so conflicted about his or her wish to dominate that A's behavior may
invite B to do the dominating (see Horowitz et al., 1997). Ideal
interacting thus requires that each partner correctly interpret
the goals of the other and behave accordingly. If the partners
misinterpret each other's goals, the behavioral reaction of one
may frustrate the goals of the other. That is why the relationship
between goals alone (or the relationship between behaviors
alone) is not an optimal predictor of satisfaction.
According to our theory, Person A is clearly gratified when
Person B's behavior complements A's goals. The theory does
not comment, however, on B's reaction to A's satisfaction. It is
possible that B would also be gratified to have complemented
A's goals. In other words, if B were able to perceive A's goal,
behaviorally complement that goal, and then observe A's satisfaction, then B, too, might derive some satisfaction from the
match. To study this issue systematically, we would need a large
sample of dyads in which one partner's interpersonal goals were
satisfied but the other partner's were not; then we could determine whether satisfaction is symmetrical or asymmetrical. The
present sample was not large enough to allow us to answer that
question.
The traditional view of complementarity needs to be modified
in another way as well. It suggests that satisfied partners judge
each other to be complementary (rather than similar). For example, satisfied dominant people should describe their partners as
submissive. In both studies, however, satisfied people judged
their partners to be similar to themselves, rather than complementary; satisfied dominant people described their partners as
dominant. Apparently, judgments about the partner's characteristics were based on the participants' satisfaction with each
other, rather than vice versa: Instead of liking someone perceived
as similar, they perceived as similar someone whom they liked.
Whenever a person interacts with a partner, that person seems
to have a particular goal that can be described along the two
interpersonal dimensions of affiliation and dominance. Why
should interpersonal goals be describable along these two dimensions? According to Bowlby (1973), Bartholomew (1990;
Bartholomew & Horowitz, 1991), and Horowitz et al. (1997),
people have two important classes of mental representations
that affect their interpersonal behavior: One describes a person's
generalized image of other people; the other describes the person's generalized image of the self. A person's interpersonal
goals may reflect these generalized images. For example, if the
person's image of other people is generally negative, he or she
may seek to avoid intimacy. Likewise, if the person's image of
the self is negative (e.g., incompetent), he or she may seek the
help of others.
To summarize, we propose a reformulation of the principle
of complementarity. We argue that people differ in their preferred ways of interacting and that these preferences, or interpersonal goals, can be described along the dimensions of affiliation
and control. One characteristic of interpersonal behaviors is
that they invite, but do not necessarily elicit, complementary
behaviors—behaviors that are similar in affiliation and reciprocal in dominance. This invitation tells the partner whether his
or her goals are likely to be satisfied. Interpersonal satisfaction
occurs when a partner's behavior invites a response that is con-
601
gruent with the person's goals. Interpersonal dissatisfaction occurs when a partner's behavior invites a response that conflicts
with the person's goals. Finally, a person's satisfaction with a
partner may lead the person to perceive the partner as similar
(rather than complementary). Our reformulation thus underscores the importance of interpersonal goals rather than the
relationship between overt behaviors in determining satisfaction
and perceived similarity.
References
Alden, L. E., Wiggins, J. S., & Pincus, A. L. (1990). Construction of
circumplex scales for the Inventory of Interpersonal Problems. Journal of Personality Assessment, 55, 521-536.
Antill, J. K. (1983). Sex role complementarity versus similarity in married couples. Journal of Personality and Social Psychology, 45, 145155.
Barry, W. A. (1970). Marriage research and conflict: An integrative
review. Psychological Bulletin, 73, 41-54.
Bartholomew, K. (1990). Avoidance of intimacy: An attachment perspective. Journal of Social and Personal Relationships, 7, 147-178.
Bartholomew, K., & Horowitz, L. M. (1991). Attachment styles among
young adults: A test of a model. Journal of Personality and Social
Psychology, 61, 226-244.
Bern, S. L. (1974). The measurement of psychological androgeny. Journal of Consulting and Clinical Psychology, 42, 155-162.
Benjamin, L. (1974). Structural analysis of social behavior. Psychological Review, 81, 392-425.
Blankenship, V., Hnat, S. M., Hess, T. G., & Brown, D. R. (1984).
Reciprocal interaction and similarity of personality attributes. Journal
of Social and Personal Relationships, 1, 415-432.
Blazer, J. A. (1963). Complementary needs and marital happiness. Marriage and Family Living, 25, 89-95.
Bluhm, C , Widiger, T, & Miele, G. (1990). Interpersonal complementarity and individual differences. Journal of Personality and Social
Psychology, 55, 464-471.
Bowlby, J. (1973). Attachment and loss, Vol. 2: Separation, New York:
Basic Books.
Bruch, M. A,, & Gilligan, J. F. (1980). Extension of Holland's theory
to assessment of marital and family interaction. American Mental
Health Counselors Association Journal, 2, 71-82.
Byrne, D. (1971). The attraction paradigm. New \brk: Academic Press.
Byrne, D., & Griffitt, W. (1969). Similarity and awareness of similarity
of personality characteristic determinants of attraction. Journal of
Experimental Research in Personality, 3, 179- 186.
Carlt, L. L., Ganley, R., & Pierce-Otay, A. (1991). Similarity and satisfaction in roommate relationships. Personality and Social Psychology
Bulletin, 17, 419-426.
Carson, R. C. (1969). Interaction concepts of personality. Chicago:
Aldine.
Clark, H. H. (1992). Arenas of language use. Chicago: University of
Chicago Press.
Clore, G. L., & Byrne, D. (1974). A reinforcement-affect model of
attraction. In T. L. Huston (Ed.), Foundations of interpersonal attraction (pp. 143-170). New York: Academic Press.
Conte, H. R., & Plutchik, R. (1981). A circumplex model for interpersonal personal traits. Journal of Personality and Social Psychology,
40, 701-711.
Deutsch, R, Sullivan, L., Sage, C , & Basile, N. (1991). The relations
among talking, liking, and similarity between friends. Personality and
Social Psychology Bulletin, 17, 406-411.
Dryer, D. C. (1993). Interpersonal goals and satisfaction with interne-
602
DRYER AND HOROWITZ
tions. Unpublished doctoral dissertation, Stanford University, Stanford, CA.
Duck, S. W. (1973). Personality similarity and friendship choice: Similarity of what, when? Journal of Personality, 41, 543-558.
Duck, S. W., & Craig, G. (1978). Personality similarity and the development of friendship. British Journal of Social and Clinical Psychology,
17, 237-242.
Estroff, S. D., & Nowicki, S., Jr. (1992). Interpersonal complementarity,
gender of interactants, and performance on puzzle and word tasks.
Personality and Social Psychology Bulletin, 18, 351-356.
Eysenck, H. L, & Wakefield, J. A. (1981). Psychological factors as
predictors of marital satisfaction. Advances in Behaviour Research
and Therapy, 3, 151-192.
Ford, M. E., & Nichols, C. W. (1991). Using goal assessments to identify motivational patterns and facilitate behavioral regulation and
achievement. Advances in Motivation and Achievement, 7, 51-85.
Griffitl, W. R. (1966). Interpersonal attraction as a function of selfconcept and personality similarity-dissimilarity. Journal of Personality and Social Psychology, 4, 581-584.
Griffitt, W. R. (1969). Personality similarity and self-concept as determinants of interpersonal attraction. Journal of Social Psychology, 78,
137-146.
Gunman, M. B. (1991). Evaluating the interpersonalness of personality
scales. Personality and Social Psychology Bulletin, 17, 670-677.
Hodges, L. A., & Byrne, D. (1972). Verbal dogmatism as a potentiator
of intolerance. Journal of Personality and Social Psychology, 21,
312-317.
Horney, K. (1945). Our inner conflicts. New York: Norton.
Horowitz, L. M. (1979). On the cognitive structure of interpersonal
problems treated in psychotherapy. Journal of Consulting and Clinical
Psychology, 47, 5-15.
Horowitz, L. M., Dryer, D. C , & Krasnoperova, E. N. (1997). The circumplex structure of interpersonal problems. In R. Plutchik & H. R.
Conte (Eds.), Circumplex models of personality and emotions. Washington, DC: American Psychological Association.
Horowitz, L. M., Inouye, D., & Siegelman, E. Y. (1979). On averaging
judges' ratings to increase their correlation with an external criterion.
Journal of Consulting and Clinical Psychology, 47, 453-458.
Horowitz, L. M , Locke, K. D., Morse. M. B., Waikar, S. V, Dryer, D. C ,
Tarnow, E., & Ghannam, J. (1991). Self-derogations and the interpersonal theory. Journal of Personality and Social Psychology, 61, 6 8 79.
Horowitz, L. M., Rosenberg, S. E., Baer, B. A., Ureno, G., & Villasenor,
V S . (1988). Inventory of interpersonal problems: Psychometric
properties and clinical applications. Journal of Consulting and Clinical Psychology, 56, 885-892.
Horowitz, L. M., Rosenberg, S. H., & Bartholomew, K. (1993). Interpersonal problems, attachment styles, and outcome in brief dynamic psychotherapy. Journal of Consulting and Clinical Psychology, 61, 5 4 9 560.
Hymel, S.. & Woody, E. (1991, April). Friends versus non-friends;
Perceptions of similarity across self, teacher, and peers. Paper presented at the biennial meeting of the Society for Research in Child
Development, Seattle, Washington.
Jellison, J. M., & Zeissct, P. T (1969). Attraction as a function of the
commonality and desirability of a trait shared by others. Journal of
Personality and Social Psychology, 11, 115-120.
John, O. P. (1990). The big-five factor taxonomy: Dimensions of personality in the natural language and questionnaires. In L. A. Pervin (Ed.),
Handbook of personality: Theory and research (pp. 66-100). New
York: Guilford Press.
Kenny, D. A. (1994). Interpersonal perception; A social relations analysis. New York: Guilford Press.
Kiesler, D. J. (1983). The 1982 interpersonal circle: A taxonomy for
complementarity in human transactions. Psychological Review, 90,
185-214.
Kiesler, D. J. (1996). Contemporary interpersonal theory and research.
New York: Wiley.
Kraemer, H. C , & Jacklin, C. N. (1979). Statistical analysis of dyadic
social behavior. Psychological Bulletin, 86, 217-224.
Krokoff, L. J., Gottman, J. M., & Roy, A. K. (1988). Blue-collar marital
interactions and a companionate philosophy of marriage. Journal of
Personal and Social Relationships, 5, 201-222.
Lafferty, J., & Eady, P. (1974). The desert survival problem. Plymouth,
MI: Experimental Learning Methods.
Leary, T. F. (1957). Interpersonal diagnosis of personality. New York:
Ronald Press.
Locke, K.D., & Horowitz, L.M. (1990). Satisfaction in interpersonal
interactions as a function of similarity in level of dysphoria. Journal
of Personality and Social Psychology, 58, 823-831.
Malle, B. F., & Horowitz, L. M. (1995). The puzzle of negative selfviews: An explanation using the schema concept. Journal of Personality and Social Psychology, 68, 470-484.
McCrae, R. R., & Costa, P.T., Jr. (1989). The structure of interpersonal
traits: Wiggins1 circumplex and the five-factor model. Journal of
Personality and Social Psychology, 56, 586-595.
McLeod, J., & Chaffee, S. (1973). Interpersonal approaches to communication research. American Behavioral Scientist. 16, 469-499.
McNemar, Q. (1962). Psychological statistics (3rd ed.). New York:
Wiley.
Moon, Y. (1996). Similarity effects in human computer interaction:
Effects of user personality, computer personality, and user control
on attraction and attributions of responsibility. Unpublished doctoral
dissertation, Stanford University, Stanford, CA.
Morell, M. A., Twillman, R. K., & Sullaway, M. E. (1989). Would a
Type A date another Type A?: Influence of behavior type and personal
attributes in the selection of dating partners. Journal of Applied Social
Psychology, 19, 918-931.
Murstein, B. L. (1961). The complementary need hypothesis in newlyweds and middle-aged couples. Journal of Abnormal and Social Psychology, 63, 194-197.
Novak, D. W., & Lerner, M. J. (1968). Rejection as a consequence of
perceived similarity. Journal of Personality and Social Psychology,
9, 147-152.
Nowicki, S., Jr., & Manheim, S. (1991). Interpersonal complementarity
and time of interaction in female relationships. Journal of Research
in Personality, 25, 322-333.
Orford, J. (1986). The rules of interpersonal complementarity: Does
hostility beget hostility and dominance, submission? Psychological
Review, 93, 365-377.
Reagor, P. A., & Clore, G. L. (1970). Attraction, text anxiety, and similarity-dissimilarity of test performance. Psychonomic Science, 18,
219-220.
Richard, L. S., Wakefield, J. A., & Lewak, R. (1990). Similarity of
personality variables as predictors of marital satisfaction: A Minnesota
Multiphasic Personality Inventory (MMPI) item analysis. Personality
and Individual Differences, fl, 39-43.
Schachter, S. (1959). The psychology of affiliation. Stanford, CA: Stanford University Press.
Strong, S. R., & Hills, H. I. (1986). Interpersonal Communication Rating Scale. Richmond, VA: Virginia Commonwealth University.
Strong, S. R., Hills, H. I., Kilmartin, C. T, De Vries, H., Lanier, K.,
Nelson, B. N., Strickland, D., & Meyer, C. W. III. (1988). The dynamic relations among interpersonal behaviors: A test of complementarity and anticomplementarity. Journal of Personality and Social
Psychology, 54, 798-810.
INTERPERSONAL COMPLEMENTARITY VERSUS SIMILARITY
Sullivan, H. S. (1953). The interpersonal theory of psychiatry. New
York: W. W. Norton.
Tharp, R. G. (1963). Psychological patterning in marriage. Psychologi-
cal Bulletin, 60, 97-117.
Wiggins, J.D., Moody, A. D., & Lederer, D. A. (1983). Personality
typologies related to marital satisfaction. American Mental Health
Counselors Association Journal, 5r 169—178.
Wiggins, J. S. (1979). A psychological taxonomy of trait-descriptive
terms: The interpersonal domain. Journal of Personality and Social
Psychology, 33, 409-420.
Wiggins, J. S. (1982). Circumplex models of interpersonal behavior in
603
clinical psychology. In P. C. Kendall & J. N. Butcher (Eds.), Handbook of research methods in clinical psychology (pp. 183-221). New
York: Wiley.
Wiggins, J. S., & Broughton, R. (1985). The interpersonal circle: A
structural model for the integration of personality research. In R.
Hogan & W. Jones (Eds.), Perspectives in personality: Theory, measurement, and interpersonal dynamics (pp. 1-47). Greenwich, CT
JAI Press.
Wish, M., Deutsch, M., & Kaplan, S. J. (1976). Perceived dimensions
of interpersonal domain. Journal of Personality and Social Psychol-
ogy, 33, 409-420.
Appendix
Subscales of the Interpersonal Goals Inventory
When I am working on a task with someone, it is important to
me to:
PA: Dominant Goals
Be self-confident.
Be firm when I need to be.
Say " n o " to the other person when appropriate.
Be aggressive when the situation calls for it.
BC: Hostile Dominant Goals
Not try to please the other person too much.
Not be too gullible.
Be assertive with the other person.
Be aggressive without worrying about hurting the other person's
feelings.
DE: Hostile Goals
Not trust the other person too much.
Not be overly generous in helping the other person.
Not put the other person's needs before my own.
Work with the person in a way that supports my own interests.
FG: Hostile Submissive Goals
Not be noticed too much.
Keep some things private from the other person.
Not open up to my partner too much.
Not tell personal things to my partner.
HI: Submissive Goals
Not fight with the other person too much.
Not be too aggressive with the other person.
Not argue with the other person too much.
Allow the other person to take control.
JK: Friendly Submissive Goals
Put the other person's needs before my own.
Not be too independent.
Not be too suspicious of the other person.
Work with the other person in a way that protects or supports the
other person's interests.
LM: Friendly Goals
Share openly my thoughts and ideas.
Not keep the other person at a distance too much.
Be supportive of the other person's goals.
Not be too cold.
NO: Friendly Dominant Goals
Let the other person know when I am angry.
Confront the other person with problems that come up.
Express my feelings to the other person directly.
Let the other person know what I want.
Received June 19, 1995
Revision received August 7, 1996
Accepted August 28, 1996 •